UEmbed

UEmbed is a decoder-only multimodal embedding model that produces both dense embeddings and SPLADE-style sparse lexical embeddings from a single causal forward pass. It supports text, image, video, and mixed-modal inputs for retrieval, multimodal search, and visual-document retrieval.

From the model card

3 variants

Sources